Limnol. Oceanogr., 46(5), 2001, 1077–1090 ᭧ 2001, by the American Society of Limnology and Oceanography, Inc.

Large-scale patterns in seagrass (Thalassia testudinum) demographics in south Florida

Bradley J. Peterson1 and James W. Fourqurean Department of Biological Sciences and Southeast Environmental Research Center, Florida International University, University Park, Miami, Florida 33199

Abstract An examination of the population age structure of 118 spatially separated subpopulations of Thalassia testudinum over the extent of the Florida Keys National Marine Sanctuary (FKNMS) during a 2-yr period revealed significant spatial variation in short shoot (SS) demographic characteristics and population dynamics. Shoot age was determined for 12,031 SS. The number of leaf scars on individual shoots was converted to shoot age by use of observed seasonally and spatially variable leaf emergence rates. The yearly mean leaf emergence rate was 0.0295 Ϯ 0.0128 leaves SSϪ1 dϪ1 (Ϯ1 SD), and the median age of counted shoots was ϳ5 yr. A significant relationship between asexual reproductive effort and gross recruitment of SS into the populations (r2 ϭ 0.15, P ϭ 0.001) and between mortality of SS and gross recruitment (r2 ϭ 0.72, P Ͻ 0.001) existed. Thus, the greatest risk of mortality occurred in areas where gross recruitment was highest. The net population growth for T. testudinum within the boundaries of FKNMS was stable (mean ϭϪ0.006 Ϯ 0.089 yrϪ1). However, areas within FKNMS fluctuated between positive and negative net growth rates (Ϫ0.20–0.50 yrϪ1). The power of such large-scale observations is the ability to identify areas of management concern and to frame questions that address the controlling mechanisms that influence these regions of fluctuating population growth.

There is a general agreement that anthropogenic alteration Quammen and Onuf 1993). The time course of changes in of the environment has influenced most areas on earth. Shal- seagrass communities caused by eutrophication can be de- low marine ecosystems, including seagrass beds, have been cadal (Fourqurean et al. 1995), resulting in slight year-to- no exception. Despite the recognition of seagrass beds as year changes in the ecosystem. Unfortunately, mapping some of the world’s most productive and valuable ecosys- methods generally suffer from lack of precision because of tems, anthropogenic losses of seagrass beds continue at an spatial and seasonal heterogeneity and sampling error; for alarming rate (Short and Wyllie-Echeverria 1996). Although this reason, mapping techniques generally require long in- anthropogenic seagrass losses have been attributed to many tervals to detect changes and necessitate repeated assess- causes (e.g., dredging, filling, oil spills, and water diversion), ments of areas to ascertain trends. Recently, Duarte et al. most such losses result from human inputs of nutrients and (1994) have advocated the use of reconstructive population sediment into the coastal zone (Short and Wyllie-Echeverria demographic methods to detect population growth trends. 1996). The mechanism of this eutrophication-associated sea- These methods rely on the ability to age individual short grass loss is clear: nutrients cause a shift in plant species shoots (SS) of seagrasses by counting leaf scars and applying dominance from slow-growing seagrasses to faster-growing a plastochron interval (PI, Erickson and Michelini 1957). competitors and, in turn, to epiphytes, macroalgae and phy- The age-frequency distribution of SS is a reflection of re- toplankton (Duarte 1995). The extensive losses of seagrass cruitment and mortality of individual SS to the seagrass pop- beds in the past few decades have focused attention on the ulation. A potential advantage of this approach is that esti- need for fast, reliable methods and tools to evaluate the ex- mates of recruitment and mortality can be obtained from a pansion or decline of seagrass populations. single sampling event. Previous examinations of seagrass Traditionally, seagrass distribution and health have been population growth dynamics that used the PI and reconstruc- assessed by long-term mapping surveys with concurrent es- tion method have consisted primarily of censuses on small timates of biomass and density (e.g., Orth and Moore 1983; spatial scales (0.1–1 km; e.g., Gallegos et al. 1993; Duarte et al. 1994; Durako 1994; Jensen et al. 1996). It is possible, 1 Corresponding author (Petersob@fiu.edu). however, that the reduced sampling effort of the reconstruc- Acknowledgments tive techniques compared with repeated mapping may permit Many people provided field and laboratory support, including regional-scale (hundreds of km), multiyear observations on Carlos Barrosso, Braxton Davis, Cassie Furst, Brian Machovina, seagrass population growth dynamics (e.g., Marba` et al. Leanne Miller, Craig Rose, and Alan Willsie. Special thanks to Cap- 1996). tian Dave Ward and the R/V Magic. The application of these techniques to populations of sea- Funding for this project was provided by grant X994620-94-5 grass SS has not been generally accepted (see Durako and awarded to J.W. Fourqurean, M.J. Durako, and J.C. Zieman from Duarte 1997; Jensen et al. 1997). Jensen et al. (1996) argue the U.S. Environmental Protection Agency as part of the Florida that the required assumptions of constant age-specific mor- Keys National Marine Sanctuary Water Quality Protection Program tality and recruitment rates are untenable. They also express and a Tropical Biology Post-doctoral Fellowship award to B.J.P. from Florida International University. concern that durations of PIs can exhibit considerable spatial This is contribution number 142 of the Southeast Environmental and temporal variation resulting from environmental influ- Research Center and contribution number 31 of the Tropical Biol- ences. Recently, Kaldy et al. (1999) reiterated this concern ogy Program at FIU. and discouraged the use of the plastochron method for con- 1077 1078 Peterson and Fourqurean

Fig. 1. Site locations of T. testudinum SS demographic samples for 1996 and 1997. struction of age-frequency distributions of Thalassia testu- extend from the southern tip of Key Biscayne to include the dinum, because this species violated the assumption of equal Dry Tortugas. For the most part, these 9,500 km2 support elapsed time between successive leaf formation. Kaldy et al. seagrass communities; Fourqurean et al. (in press) document (1999) further proposed that this bias in leaf formation mea- that there are Ͼ15,000 km2 of seagrass beds in the south surements leads to the commonly reported seagrass age-fre- Florida region. quency distributions with too few young shoots to account for the older shoots in the populations. Leaf emergence rates—The rate at which new leaves were In this study, the potential of the PI–reconstructive de- produced by SS was determined by leaf punching all SS mographic techniques to regional assessments of the state of within six 0.04 m2 quadrants quarterly at 28 permanent sites seagrass populations was explored. In addition, we attempted within FKNMS in 1996 and 1997 (Fig. 1). These 28 sites to address the concerns raised by critics of the approach and correspond to a concurrent program of water quality moni- reassessed the mathematical models used to derive recruit- toring and were chosen by use of a stratified-random ap- ment and mortality from the age-frequency data. Data from proach, where distance offshore and broad geographic lo- a regional, multiyear, quarterly sampling program was used cations were the strata. After periods of 9–15 d, all SS within to assess spatial and seasonal variability in PIs. Then the the quadrants were harvested. The leaf emergence rate (LER, temporally and spatially variable estimates were applied to in new leaves SSϪ1 dϪ1) was calculated by dividing the num- collections of SS of T. testudinum from throughout the re- ber of new leaves produced in the incubation period by the gion, to estimate population growth parameters for the sub- total number of leaf-punched SS, then dividing by the time populations. Although questions remain about the applica- interval between marking and harvest. Note that LER is the bility of the reconstruction method as a tool for predicting reciprocal of PI, or the number of days between formation change in seagrass populations, we believe that this tech- of successive leaves on a SS (Patriquin 1973; Brouns 1985). nique can be used with caution to examine regional differ- To evaluate the effects of seasonal variation on the estimates ences in population dynamics of seagrass beds. of LER, a sine-wave function (Eq. 1) was fitted to the data on leaf emergence rates acquired from the permanent quar- Methods terly sites by use of nonlinear least-squares regression: LER ϭ LER ϩ Amp[sin(DOY ϩ ␾ )], (1) The Florida Keys National Marine Sanctuary (FKNMS) consists of ϳ9,500 km2 of coastal and oceanic waters that where LER is the mean annual PI, Amp is the amplitude of Variability in seagrass demographics 1079

␾ the sine wave, DOY is the day of year in radians, and is the estimated number of SS in the youngest age class, Nt is the phase angle. This equation was fitted to LER rather than the number of SS in the tth age class, and M is the per capita to PI because LER approaches 0 in the winter months; hence mortality rate. Every age class contributed to the estimate of PI approached ϱ, which caused instability in the nonlinear M. This calculation assumes (1) age-independent mortality, regression. (2) constant annual SS mortality, and (3) constant annual recruitment rates. All of the SS age classes were used to Population demographics: determining age structure— estimate M instead of the numbers in cohort peaks, as has During the summers of 1996 and 1997, samples of T. tes- been used elsewhere (Durako 1994). We feel that this avoids tudinum to be used in demographic analyses were obtained biasing the estimate of M by only sampling individuals re- from 118 randomly selected sites within the FKNMS (Fig. cruited to the population during cohort peaks; our method 1). Samples from both years were quasi evenly distributed also allows for an analysis of the residuals to determine co- throughout our study area. Populations of SS were collected horts, although this was not applied in the present study. from each site by excavation of a sod, ϳ1m2, that contained Ͼ100 SS. Care was taken to excavate the entire root-rhizome Population growth rates: per capita recruitment—Gross complex within a sod, but SS sometimes broke from the recruitment was estimated from the age-frequency distribu- rhizomes. There is some possibility that inclusion of these tions by first estimating the number of new SS produced in broken SS in our analyses would lead to the underestimation the population within the last year and dividing by the total of the age of those SS. Most of the time, however, this break number of SS. For this to properly estimate gross recruit- occurred at or very near the rhizome, and we rarely observed ment, mortality of SS aged Ͻ1 yr must be accounted for;

broken SS still attached to the rhizome, so we felt that SS hence, we calculated gross recruitment (Rgross)as

not attached to the rhizome could be included in our age- k ϩ frequency estimations. Not including broken SS in the sam- ͸ e(ln njjMt ) ple could also lead to a bias, because we noted that older, jϭ1 R ϭ , (3) longer SS were more likely to break off their rhizomes than gross N younger, shorter SS. The sods were placed in mesh bags, where the sediment was gently washed from the root-rhi- where n is the number of SS in the jth age class, t is the age Ͻ zome complex. Samples were returned to the laboratory for of the jth age class, k is the oldest age class 1 yr old, M processing. The number of leaves produced over the lifespan is the per capita mortality rate, and N is the total number of of each SS in a sample was determined by counting leaf SS in all age classes. Previous estimates of gross recruitment scars and extant green leaves. SS age was estimated by the from seagrass age-frequency distributions (e.g., Duarte et al. number of leaves produced by a SS, scaled by the site-spe- 1994; Durako 1994) have not taken mortality of the aged SS Ͻ cific annual leaf production rates (Patriquin 1973; Duarte et 1 yr into account. Note that this calculation of gross re- al. 1994). This site-specific leaf production rate was calcu- cruitment is also dependent on the assumptions of age-in- lated by interpolating the site-specific LER from the 28 per- dependent mortality and constant interannual mortality. manent stations by use of a kriging algorithm. Gross SS recruitment and mortality rates were used to calculate population growth as net annual new SS produc- Population demographics: reproductive effort—During tion, 1997, the number of flowers, fruits, and rhizome apical mer- ϭ Ϫ Pnet Rgross M, (4) istems in each demographic sample was recorded. No at- Ͼ tempt was made to determine sex ratio. The percentage of which predicts whether the population is expanding (Rgross Ͻ ϭ SS carrying flowers or fruits was determined, to examine the M), declining (Rgross M), or in steady state (Rgross M). sexual reproductive effort. With this method, there is an in- In assessing spatial pattern in Rgross, M, and Pnet, we treated herent underestimation of sexual reproduction, because it samples collected in both sampling years as part of the same only records the fruit and flowers present at the time of sam- statistical population; in effect, this is explicitly recognizing pling; however, this was an attempt to discover any patterns the assumptions of the demographic models that Rgross and of sexual reproductive effort in space. Because sampling oc- M are constant through time. Interannual variation in Rgross, curred over a short time interval during the summer of 1997, M, and Pnet will be explored in a future article. we believe that the bias was uniformly applied across all sites. In addition, the number of horizontal rhizome apicals Results as a percentage of the total number of SS were determined to examine the asexual reproductive effort at each site. Leaf emergence rates—LER was measured in 210 out of a possible 224 times (28 sites ϫ 8 sampling intervals); the Population growth rates: per capita mortality—Age-fre- differential was due to storm damage, lost data, and minor quency distributions of SS were constructed for each de- delays in establishing some sites. Measured LER ranged mographic site. Instantaneous mortality rate, derived from from a minimum of 0.0016 to a maximum of 0.0650 leaves the number of SS in each age class, was estimated by fitting SSϪ1 dϪ1, which translates into PI estimates of between 15.4 the equation and 611 d leafϪ1. The mean LER was 0.0295 Ϯ 0.0128 leaves SSϪ1 dϪ1 (Ϯ1 SD). The frequency distribution of PI N ϭ N eϪMt (2) t o estimates was skewed and failed tests of normality; the me- Ϫ1 to the age-frequency distributions at each site, where No is dian PI was 34.7 d leaf . 1080 Peterson and Fourqurean

dence intervals (CI) for the parameters were 0.0218–0.0335 leaves SSϪ1 dϪ1 for LER, 0.0013–0.0187 leaves SSϪ1 dϪ1 for Amp, and 3.873–5.500 for ␾. The phase angle indicates the timing in the annual peak and nadir in LER; adding 3␲/2 to the value in radians and converting from radians to calendar dates yields an estimate of the timing of the peak production of new leaves between 14 May and 17 August and the timing of the minimum production between 13 November and 15 February.

Population demographics: determining age structure—In general, the 118 sites sampled for the age structure of T. testudinum SS were found to have populations that were skewed toward young individuals, with apparent cohort peaks approximately every 10 leaf scars (Fig. 4A,B). There were large ranges, however, in the predominance of young SS among the sites. The median number of leaf scars on SS from a sample ranged from 20 to 284 scars SSϪ1, with a mean for all 118 sites of 53.5 Ϯ 26.3 scars SSϪ1 (Fig. 5A). There was similar variability in the maximum number of leaf scars found on an individual SS at a site, with a range of 39–566 scars SSϪ1and a mean of 103.6 Ϯ 52.5 scars SSϪ1 (Fig. 5B). The interpolated fields of LER calculated from the mea- sured LER from the 28 permanent sites (Fig. 3) were used to estimate LER at all of the 118 demographics sites. These average annual rates of leaf emergence were used to convert the leaf scar count data from each demographic site into SS ages (Fig. 4C,D). The average median and maximum age of SS at a site were 5.0 Ϯ 2.6 and 9.7 Ϯ 5.1 yr, respectively Fig. 2. LER for T. testudinum. (A) Example data from one of (Fig. 5C,D). Spatial plots of median SS age revealed that the permanent stations (Sta. 314, see Fig. 1 for location). (B) The the oldest subpopulations of T. testudinum occurred on the mean LER from all permanent stations (nominally N ϭ 28). Error Atlantic Ocean side of the Florida Keys throughout the ex- bars are 1 SD. Solid line represents a nonlinear least-squares fit of tent of FKNMS, whereas subpopulations on the Gulf of the sine model (Eq. 1) to the data. The bottom X axis is time of Mexico side of the Florida Keys had younger mean SS age year in radians; the top X axis converts radians to months. The dashed horizontal line is LER from the model fit. (data not shown). Thus, the oldest populations existed off- shore, and SS age decreased along a gradient from offshore to Florida Bay. The sine model (Eq. 1) described the site-specific tem- poral pattern of LER quite well (see example data set, Fig. Population demographics: reproductive effort—An ex- 2A). An average of 38% of the variance in LER at a site amination of the horizontal rhizome apical meristems per SS was described by the model, and in all 28 instances the pa- revealed a maximum level of asexual reproductive effort at rameter estimates for LER and Amp were significantly dif- 25%, with a mean of 8.3% Ϯ 5.5 (1 SD) over the extent of ferent from 0 at P Յ 0.10. The mean of the estimated LER the FKNMS (insert, Fig. 6A). In contrast, the greatest level from the 28 sites was 0.0296 Ϯ 0.0046 leaves SSϪ1 dϪ1, with of sexual reproductive effort was 25%, with a mean of 2.3% a range of 0.0221–0.0406 leaves SSϪ1 dϪ1. Mean annual leaf Ϯ 5.0% (1 SD). The frequency histogram for SS that bore emergence rates were normally distributed about the mean fruits or flowers was highly skewed, revealing that low levels (inset, Fig. 3). The mean estimated Amp was 0.0098 Ϯ of sexual reproduction occur throughout the extent of 0.0050 leaves SSϪ1 dϪ1, with a range of 0.0018–0.0183 FKNMS. Thus, the average level of asexual reproductive leaves SSϪ1 dϪ1. There was a bimodal distribution of Amp effort in the FKNMS is four times greater than that of sexual estimates (inset, Fig. 3). There was spatial heterogeneity in reproductive effort (insert, Fig. 6B). Spatially, the popula- the estimates of both model parameters (Fig. 3), with no tions with the greatest asexual reproductive effort occurred clear relationship between LER and Amp or between these predominately on the Gulf of Mexico side of the Florida parameters and obvious physical factors, such as light avail- Keys and adjacent to the northeastern Florida Keys ocean- ability and depth (nonsignificant linear regressions; data not side (Fig. 6A), whereas the populations with the highest sex- shown). ual reproductive effort occurred on the ocean side of the As an alternative method of describing the general pattern Florida Keys, with four spatially coherent regions having in LER, the sine model (Eq. 1) was also fitted to the mean levels Ͼ10%, located offshore of Key Largo, Lower Matec- LER at all sites for each sample time (Fig. 2B). When the umbe, Summerland, and west of Key West (Fig. 6B). The model was applied to these means, asymptotic 95% confi- highest levels of asexual reproductive effort and sexual re- Variability in seagrass demographics 1081

Fig. 3. Spatial pattern in LER (top) and Amp (bottom) for T. testudinum. Frequency distributions for each parameter are in the insets, N ϭ 28. Station locations are indicated by small crosses. Note that the contour maps indicate the spatial extent of the study, not the limits of T. testudinum distribution in south Florida. 1082 Peterson and Fourqurean

Fig. 4. Leaf scar and SS age frequency histograms from a high and low mortality site in FKNMS. (A) Leaf scar frequency histogram for site D39. (B) Leaf scar frequency histogram for site 305. (C) Frequency histogram of SS age for site D30. (D) Frequency histogram of SS age for site 305. productive effort occurred in different spatial regions of shore to Florida Bay, with a peak in mortality north of Mar-

FKNMS. athon (Fig. 7B). Thus, the greatest levels of Rgross occurred in areas of FKNMS that experience the greatest levels of Population growth rates: per capita recruitment and mor- mortality (Fig. 9). tality—Comparisons of the population growth parameters in- dicate significant demographic differences across the span of Regional pattern in predicted net population growth—

FKNMS. Estimates of Rgross for the 118 subpopulations of T. Predicted net population growth (Pnet in Eq. 4) for the 118 testudinum sampled ranged from 0.01 to 1.30 yrϪ1 (insert, subpopulations of T. testudinum from the FKNMS varied Fig. 7A), with a mean of 0.17Ϯ 0.12 yrϪ1. Spatially, the from Ϫ0.20 to 0.50 yrϪ1, with a mean (Ϫ0.006 Ϯ 0.089 yrϪ1) highest Rgross was found within Florida Bay north of Mara- that was not significantly different from 0 (inset, Fig. 10). thon (Fig. 7A). In general, annual recruitment increased from This indicates that, on the whole, the T. testudinum popu- offshore to nearshore oceanside of FKNMS. Gross recruit- lation in the region is in steady state, but individual subpop- ment and asexual reproductive effort were significantly pos- ulations are dynamic. Two distinct regions of net population itively correlated (Fig. 8), but Rgross and sexual reproductive growth and decline were evident (Fig. 10). Our analyses pre- effort were not correlated. dicted expanding subpopulations of T. testudinum offshore

As with Rgross, mortality of SS also ranged from 0.01 to between Marathon and Bahia Honda and in the far western 1.30 yrϪ1, with a mean of 0.18Ϯ 0.09 yrϪ1. In a pattern extent of Florida Bay. Furthermore, we predicted that T. tes- similar to that of Rgross, mortality rates increased from off- tudinum subpopulations were declining offshore of Planta- Variability in seagrass demographics 1083

Fig. 5. Median and maximum leaf scar and SS age frequency histograms. (A) Median leaf scar frequency histogram. (B) Maximum leaf scar frequency histogram. (C) Median SS age frequency histogram. (D) Maximum SS age frequency histogram. tion and Upper Matecumbe Keys and within Florida Bay 0.91, respectively). There was a positive correlation between north of Marathon and Bahia Honda. However, these pre- depth and the sexual reproductive effort (P ϭ 0.03), which dictions of net population change generated from our anal- indicates that those T. testudinum meadows at deeper sites ysis of the age-frequency distributions of SS within subpop- had greater numbers of flowers or fruits. In addition, there ulations remain to be tested. was a positive correlation between maximum leaf scars and mean S. filiforme standing crop, as well as between median Correlation of environmental parameters with demo- leaf scars and mean S. filiforme standing crop. This indicates graphics data—At 22 of the permanent sites, several envi- that the oldest populations of T. testudinum in these per- ronmental parameters measured for another study (Fourqur- manent sites occur in areas where S. filiforme standing crops ean et al. 2001) could be correlated with the demographics are greatest. data presented here. These parameters included depth, light reaching the bottom, seasonality of T. testudinum areal pro- Discussion ductivity, seasonality of T. testudinum standing crop, mean Syringodium filiforme standing crop, and seasonality of S. The ability to determine the age of an organism has filiforme standing crop. There was no correlation between proved useful in elucidating its ecology and population dy- depth and mortality, recruitment, or net population growth namics. A population age structure is simultaneously the out- (P ϭ 0.08, 0.48, and 0.74, respectively). Nor was there a come of past demographic events and an indication of its correlation between light reaching the bottom and mortality, demographic future. The current population age structure re- recruitment, or net population growth (P ϭ 0.15, 0.28, and flects previous temporal variation in recruitment and mor- 1084 Peterson and Fourqurean

Fig. 6. Spatial pattern in asexual reproductive output (top) and sexually active SS (bottom) throughout the extent of FKNMS. Frequency histograms for asexual reproductive output (insert, top) and sexually active SS (insert, bottom) are shown. tality and can predict future population growth (Caswell sonality in leaf emergence rates and applied refinements in 1989; Bullock et al. 1996). Hence, the ability to determine the extraction of demographic parameters from age-frequen- the age of shoots of T. testudinum provides a powerful tool cy distributions. We have also used these techniques for the for clarifying population dynamics of this species (Gallegos analysis of the demographics of T. testudinum over an un- et al. 1993; Duarte et al. 1994; Durako 1994). Although precedented spatial scale and have documented large-scale techniques for determining the age of seagrass shoots have patterns in shoot demographics. been available for the past three decades (Patriquin 1973), As noted by critics of reconstructive aging techniques they have only recently been used extensively. As these tools (Jensen et al. 1996; Kaldy et al. 1999), LERs, critical to the are exercised more frequently, the techniques continue to conversion of leaf-scar information into age, are not constant improve. In the present study, we have accounted for sea- in either space or time. We documented considerable varia- Variability in seagrass demographics 1085

Fig. 7. Spatial pattern in annual recruitment (top) and mortality (bottom) over the extent of the FKNMS. Frequency histograms of recruitment estimates (insert, top) and mortality estimates (insert, bottom) at the 118 sites with the mean and 95% CI are shown. tion in LERs across our study area (Fig. 3): it is imperative ida (Fig. 2B). For comparison, we applied our sine model to that site-specific measures of LER rate be used when cal- the data on intra-annual variation in leaf emergence rates in culating the ages of SS. We also applied a novel approach Puerto Morelos, Mexico (Gallegos et al. 1993), and Laguna for incorporation of seasonal variation in LERs into an es- Madre, Texas (Kaldy et al. 1999). In Puerto Morelos, a more timate of yearly average LER (LER). We fitted a sine model tropical location than south Florida, the sine model estimated to time series of measured LERs; this allowed for calculation LER for three sites to be between 0.035 and 0.041 leaves of both a yearly mean (LER) and a measure of seasonal SSϪ1 dϪ1, with Amp varying from 0.004 to 0.017 leaves SSϪ1 variation (Amp). The amplitude of the yearly signal in the dϪ1; the amplitude of seasonal variation was 24% of the LER of T. testudinum was ϳ33% of the mean in south Flor- mean. In Laguna Madre, a less tropical location than south 1086 Peterson and Fourqurean

Fig. 8. Relationship between T. testudinum gross recruitment (yrϪ1) and asexual reproductive effort (%) at 118 sites in south Florida.

Fig. 9. Scatter plot of estimated mortality and gross recruitment for T. testudinum subpopula- tions from 118 sites in south Florida. Variability in seagrass demographics 1087

Fig. 10. Spatial plot of the predicted net population growth of T. testudinum within the boundaries of the FKNMS. Frequency histogram of the estimated net population growth for the 118 samples with mean and 95% CI is shown as the insert.

Florida, the amplitude of seasonal variation was 43% of the peak leaf emergence (Fig. 2). Summer estimates of PI for T. LER. testudinum from the literature are all quite short, with PI We propose that the use of an annual LER is an improve- generally Ͻ20 d (Table 1). In contrast, studies that have ment of the seagrass demographic and reconstruction tech- assessed PI on an annual basis produce longer estimates of nique over other estimates of LER. In the work presented PI, from 21 d in the Yucatan to 187 d in the northern Gulf here, there were no significant differences in the average of Mexico (Table 1); our estimates are within this range. On LER measured across time at a site and the estimate of LER, the other hand, there may be real biological reasons that because our sampling events were evenly distributed in time. Durako (1994) could have found a shorter PI than our esti- However, the estimates of LER from the sine model may be mates. Durako was studying the effects of the poorly un- much different than the simple average if the sampling derstood seagrass die-off in Florida Bay, and he noted that events are not evenly spaced in time. The calculation of an there was a pattern of decreasing PI concurrent with increas- annual estimate of LER incorporates the potentially consid- ing magnitude of disturbance. In effect, PI decreased in his erable temporal variation in productivity resulting from fluc- study as the seagrass canopy was thinned, which possibly tuating environmental influences, which has been expressed indicates a growth response to the thinning canopy. Indeed, previously by some investigators as an obstacle to applying sparse SS that survived drastic canopy thinning had PI val- reconstructive techniques (Jensen et al. 1996; Kaldy et al. 1999). ues as short as 4 d (Durako 1994). It is also possible that The average yearly mean LER found in this study predicts our sampling regime missed peak times of leaf emergence aPIof34dforT. testudinum in south Florida. This value because of lowered leaf emergence as a consequence of sum- is high in comparison with the 13–18 d found by Durako mer sexual reproduction (as documented by Kaldy et al. (1994) in an examination of T. testudinum SS age structure 1999). However, the low rate of sexual reproduction mea- in Florida Bay; it is also higher than the average of 15.6 d sured in this study (Fig. 6B) argues against this contention. reported in a review of the literature on this species (Duarte In addition to the use of an annual estimate of LER, we 1991). This discrepancy is most likely a result of earlier modified the gross recruitment equation to incorporate mor- estimates of PI that were based on few measurements con- tality of shoots aged Ͻ1 yr. Previously (Duarte et al. 1994), centrated during late spring and early summer, the period of gross recruitment was calculated by the equation 1088 Peterson and Fourqurean

Table 1. Comparison of plastochron interval (PI) estimates for Thalassia testudinum from a variety of locations derived from leaf marking. Some measurements were made only once, during late spring to early summer, whereas others represent yearly means.

PI (d leafϪ1 ssϪ1) Location Latitude (N) Reference Measurements made during summer 12 Bermuda 32Њ17Ј Patriquin (1973) 13–18 South Florida 25Њ Durako (1994) 14–16 South Florida 25Њ Zieman (1975) 20–47 South Florida 24Њ30Ј Tomasko and Lapointe (1991) 20–27 Belize 18Њ Tomasko and Lapointe (1991) 11–20 Barbados 13Њ10Ј Patriquin (1973) Annual estimates 187 Northern Gulf of Mexico 30Њ Eleuterius (1987) 29–38 Texas 26Њ10Ј Kaldy et al. (1999) 25–50 South Florida 24Њ30Ј Present study 21–26 Yucatan, Mexico 21Њ Gallegos et al. (1993) 27–29 Yucatan, Mexico 20Њ30Ј van Tussenbroek (1995)

ϭ ͸͸Ϫ Rgrossln S total ln Stϩ1, (5) dynamics in south Florida within the boundaries of the FKNMS. where Stotal is the total number of SS in a sample and Stϩ1 is the number of SS aged Ͼ1 yr. Equation 5 does not account An attempt to correlate the estimated T. testudinuim pop- for the loss of new recruits during the year into an estimate ulation demographic parameters to environmental data re- of gross recruitment. Obviously, mortality occurs within vealed that there were no statistically significant relation- shoots Ͻ1 yr old. We propose that our modified gross re- ships with depth or light reaching the bottom. This lack of cruitment equation (Eq. 3) more accurately reflects what oc- correlation suggests that the amount of light reaching the curs within natural seagrass populations. Comparison of es- bottom is not a primary controller of recruitment and mor- timates of gross recruitment generated by Eq. 3 with those tality of T. testudinum in south Florida. It is possible some calculated by Eq. 5 revealed that Eq. 3 leads to a significant as-yet-unexamined physical parameter may correlate with SS increase in the estimation of gross recruitment. On the basis mortality and recruitment. For example, seasonality in the of the 118 subpopulations of T. testudinum in this study, Eq. productivity of T. testudinum appears to be related to sea- 3 resulted in an increase in estimated gross recruitment that sonality in water temperature in south Florida (Fourqurean ranged from 1% to 56%, with a mean increase of 13%. Of et al. 2001). Furthermore, small-scale physical disturbances particular interest is that the difference between the two (e.g., ‘‘blowouts,’’ Patriquin 1975) can create variability in the species composition and possibly demographics of sea- methods increases as Rgross increases. Therefore, use of Eq. 5 results in greater underestimations of gross recruitment in grass beds; this could obscure relationships between demo- subpopulations that experience higher levels of recruitment graphics and other environmental driving variables. In ad- and mortality. Although questions remain about the validity dition, biotic interactions may attribute to the spatial of the assumptions required by the reconstructive technique, variability of T. testudinum population growth in south Flor- our modifications will improve the estimates of shoot age, ida. Several recent studies have demonstrated that herbivory annual mortality, annual recruitment, and, consequently, net of seagrass can increase shoot density and seagrass produc- population growth. tivity in the northern Gulf of Mexico and south Florida (Val- Analysis of the population age structure of 118 subpop- entine et al. 1997, 2000), or herbivory can totally disrupt the ulations of T. testudinum throughout the extent of FKNMS previous age structure of the population (Rose et al. 1999). over a 2-yr period revealed significant spatial variation in All of these possible factors have yet to be tested. Currently, SS demographic characteristics and in the predicted popu- we are examining whether the estimates of recruitment and lation dynamics. There was a spatial overlap of the popu- mortality calculated in the present study accurately reflect lations with the highest asexual reproductive effort (Fig. 6A) the population growth dynamics of T. testudinum in subse- quent years. and high Rgross (Fig. 7A) in Florida Bay and the Upper Keys but not throughout the rest of FKNMS. Similarly, a less dra- Discrete areas of high and low mortality and high and low matic overlap of populations with high sexual reproductive recruitment were observed in FKNMS. The spatial patterns effort with high mortality existed in the Upper Keys and one of high gross recruitment and high mortality were nearly region of the Florida Bay. A significant correlation between congruent. Thus, the greatest risk of mortality occurred in populations with high asexual reproductive effort and Rgross areas where gross recruitment was highest. This pattern has (Fig. 8) and between populations with Rgross and high mor- been shown elsewhere for perennial and annual terrestrial tality (Fig. 9) existed over the study area. Throughout the plants (Bullock et al. 1996; Noble and Dirzo 1997) and is extent of FKNMS, sexual reproductive effort did not cor- likely a result of intraspecific competition. It is possible that 2 ϭϪ ϭ relate to Rgross (r 0.012, P 0.908) or to asexual re- the increased density due to new recruits increases the mor- productive effort (r2 ϭ 0.021, P ϭ 0.102). Thus, we con- tality rate as a result of reduced light levels caused by shad- clude that asexual vegetative growth drives recruitment ing or reduced per capita nutrient availability. Alternatively, Variability in seagrass demographics 1089 the reduced density caused by mortality may enhance re- symposium on subtropical-tropical seagrasses of the south- cruitment. However, regardless of the mechanism, it is note- eastern United States. Florida Marine Research Publication No. worthy that estimating gross recruitment with Eq. 5 increas- 42. Florida Department of Marine Research. ingly underestimates recruitment as recruitment increases ERICKSON,R.O.,AND F. J. MICHELINI. 1957. The plastochron index. and that, in FKNMS, the gross recruitment was positively Am. J. Bot. 44: 297–305. FOURQUREAN J.W.,M.J.DURAKO,M.O.HALL, AND L. N. HEFTY. correlated with mortality. Thus, the use of Eq. 5 would sub- In press. Seagrass distribution in south Florida: A multi-agency stantially underestimate Pnet in these areas and predict neg- coordinated monitoring program. In J. W. Porter and K .G. ative net population growth in very dynamic seagrass sub- Porter [eds.], Linkages between ecosystems in the south Flor- populations. ida hydroscape: The river of grass continues. CRC. The unprecedented scale of examination undertaken in the ,G.V.N.POWELL,W.J.KENWORTHY, AND J. C. ZIEMAN. present study allowed regional observations of T. testudinum 1995. The effects of long-term manipulation of nutrient supply population dynamics. Our demographic analyses indicated on competition between the seagrasses Thalassia testudinum that the net population density of T. testudinum over the and Halodule wrightii in Florida Bay. Oikos 72: 349–358. breadth of FKNMS was stable, because the mean estimated ,A.W.WILLSIE,C.D.ROSE, AND L. M. RUTTEN. 2001. Spatial and temporal pattern in seagrass community composi- Pnet was not statistically different from 0. However, age structures of populations suggests that areas within FKNMS tion and productivity in south Florida. Mar. Biol. 138: 341– 354. fluctuate between positive and negative net growth rates. GALLEGOS, M. E., M. MERINO,N.MARBA´ , AND C. M. DUARTE. Smaller observations in isolated areas might lead to opposite 1993. Biomass and dynamics of Thalassia testudinum in the conclusions about the overall state of T. testudinum popu- Mexican Caribbean: Elucidating rhizome growth. Mar. Ecol. lations in FKNMS. Prog. Ser. 95: 185–192. We believe that the changes proposed in this study for JENSEN, S. L., B. D. ROBBINS, AND S. S. BELL. 1996. Predicting estimating an annual PI and calculating gross recruitment are population decline: Seagrass demographics and the reconstruc- improvements to the reconstruction technique and should be tive technique. Mar. Ecol. Prog. Ser. 136: 267–276. considered in future studies. In addition, we suggest that the , , AND . 1997. On the use of the reconstruc- assumptions of this technique be tested experimentally. Al- tive technique: Criticisms, comments and questions. Mar. Ecol. though the validity of the assumptions used in the recon- Prog. Ser. 146: 305–309. structive technique for seagrass is still debatable and use of KALDY, J. E., N. FOWLER, AND K. H. DUNTON. 1999. Critical as- the calculated population parameters to predict future trends sessment of Thalassia testudinum (turtle grass) aging tech- niques: Implications for demographic inferences. Mar. Ecol. in populations is still largely untested, the power of such Prog. Ser. 181: 279–288. large-scale observations is in their ability to identify areas MARBA` N., C. M. DUARTE,J.CEBRIA` N,M.E.GALLEGOS,B.OL- of management concern and to shape questions about the SESEN, AND K. SAND-JENSEN. 1996. Growth and population controlling mechanisms that influence these regions of fluc- dynamics of Posidonia oceanica on the Spanish Mediterranean tuating population growth. coast elucidating seagrass decline. Mar. Ecol. Prog. Ser. 137: 203–213. References NOBLE,I.R.,AND R. DIRZO. 1997. Forests as human dominated ecosystems. Science 277: 522–525. RTH AND OORE BROUNS, J.J.W.M. 1985. The plastochron interval method for the O ,R.J., K. A. M . 1983. Chesapeake Bay: An un- study of the productivity of seagrasses: Possibilities and limi- precedented decline in submerged aquatic vegetation. Science tations. Aquat. Bot. 21: 71–88. 222: 51–53. BULLOCK, J. M., J. SILVERTOWN, AND B. C. HILL. 1996. Plant de- PATRIQUIN, D. 1973. Estimation of growth rate, production and age mographic responses to environmental variation: Distinguish- of the marine angiosperm Thalassia testudinum K. Caribb. J. ing between effects on age structure and effects on age-specific Sci. 13: 111–123. vital rates. J. Ecol. 84: 733–743. . 1975. ‘‘Migration’’ of blowouts in seagrass beds at Bar- CASWELL, H. 1989. Matrix population models. Sinauer. bados and Carricou, West Indies, and its ecological and geo- DUARTE, C. M. 1991. Allometric scaling of seagrass form and pro- logical implications. Aquat. Bot. 1: 163–189. ductivity. Mar. Ecol. Prog. Ser. 77: 289–300. QUAMMEN, M. L., AND C. P. ONUF. 1993. Laguna Madre: Seagrass . 1995. Submerged aquatic vegetation in relation to different changes continue decades after salinity reduction. Estuaries 16: nutrient regimes. Ophelia 41: 87–112. 302–310. , AND OTHERS. 1994. Reconstruction of seagrass dynamics, ROSE,C.D.,AND OTHERS. 1999. Overgrazing of a large seagrass age determinations and associated tools for the seagrass ecol- bed by the sea urchin Lytechinus variegatus in outer Florida ogist. Mar. Ecol. Prog. Ser. 107: 195–209. Bay. Mar. Ecol. Prog. Ser. 190: 211–222. DURAKO, M. J. 1994. Seagrass die-off in Florida Bay (USA): SHORT,F.T.,AND S. WYLLIE-ECHEVERRIA. 1996. Natural and hu- Changes in shoot demographic characteristics and population man-induced disturbance of seagrasses. Environ. Conserv. 23: dynamics in Thalassia testudinum. Mar. Ecol. Prog. Ser. 110: 17–27. 59–66. TOMASKO,D.A.,AND B. E. LAPOINTE. 1991. Productivity and bio- , AND C. M. DUARTE. 1997. On the use of reconstructive mass of Thalassia testudinum as related to water column nu- aging techniques for assessing seagrass demography: A critique trient availability and epiphyte levels: Field observations and of the model test of Jensen et al. (1996). Mar. Ecol. Prog. Ser. experimental studies. Mar. Ecol. Prog. Ser. 75: 9–17. 146: 297–303. VALENTINE, J. F., K. L. HECK,JR., J. BUSBY, AND D. WEBB. 1997. ELEUTERIUS, L. N. 1987. Seagrass ecology along the coasts of Al- Experimental evidence that herbivory increases shoot density abama, Louisiana and Mississippi, p. 11–24. In M. J. Durako, and productivity in a subtropical turtlegrass (Thalassia testu- R. C. Phillips, and R. R. Lewis III [eds.], Proceedings of the dinum) meadow. Oecologia 112: 193–200. 1090 Peterson and Fourqurean

, ,K.D.KIRSCH, AND D. WEBB. 2000. Seagrass her- ogy of turtle grass, Thalassia testudinum, p. 541–562. In L. E. bivory in the turtlegrass habitats of the Florida Keys. Mar. Cronin [ed.], Estuarine research, v. 1. Academic. Ecol. Prog. Ser. 200: 213–228. VAN TUSSENBROEK, B. I. 1995. Thalassia testudinum leaf dynamics Received: 6 June 2000 in a Mexican coral reef lagoon. Mar. Biol. 122: 33–40. Amended: 23 February 2001 ZIEMAN, J. C. 1975. Quantitative and dynamic aspects of the ecol- Accepted: 31 March 2001